Hybrid Rank-Based Panel Unit Root Tests

We study optimality of unit root tests in a semiparametric panel setting with both large time-series and cross-sectional dimension. While, with finite cross-sectional dimension, the limit experiments are locally asymptotically Brownian functional (LABF), we find the classical locally asymptotically normal (LAN) structure with large cross-sectional dimension. This leads to an upper bound for the power of tests using sequential asymptotics. We propose a hybrid rank-based test, based on cross-sectional ranks per time unit, that is point optimal. We corroborate our theoretical results with a Monte Carlo study.